Re: module mess in -CURRENT

[Andi Kleen]

On Thu, Nov 14, 2002 at 07:01:17PM +0100, Andrea Arcangeli wrote:
> On Thu, Nov 14, 2002 at 06:40:49PM +0100, Andi Kleen wrote:
> > > Owens' kbuild-2.5 handled it a different way and didn't need exact
> > > timings. That is especially important since nanosecond time accuracy is
> > 
> > You may not believe it, but there are projects other than the kernel
> > that do use make too.
> > 
> > > impossible if you are handling a collection of machines doing the
> > > work. NTP is accurate, but not that accurate.
> > 
> > The patch does not actually implement nanosecond resolution, but
> > jiffies resolution (1ms on 2.5), which is easily reachable with NTP.
> 
> 1msec still leave a reasonable window open IMHO. this problem would need
> sequence numbers updated atomically to be solved correctly without
> regard to the timing that may vary depending on the hardware IMHO (so
> that it could still work on a 1Thz cpu even if it's not a pratical
> matter for at least this decade).
> 

The API is ready for nanoseconds so when you feel the need you can 
convert it to using do_gettimeofday()

-Andi